Add a centralized atomic write utility and directory metadata helpers (internal/orchestrator/fs_atomic.go) and unit tests (fs_atomic_test.go). Replace ad-hoc MkdirAll/WriteFile calls in staged-apply paths (network, PBS, PVE, firewall, HA, notifications) to use ensureDirExistsWithInheritedMeta and writeFileAtomic so staged restores are applied atomically and final ownership/permissions are enforced (avoiding umask issues). Update docs: README credits release testers and docs/RESTORE_GUIDE.md and TROUBLESHOOTING.md explain atomic/staged behavior and PBS permission recovery steps. Update internal backup lock metadata (.backup.lock). Also remove the previous duplicate writeFileAtomic in restore_access_control.go in favor of the new centralized implementation.

Remove stale .backup.lock and update orchestrator tests to create isolated temp directories for backup, logs and lock files. Tests now populate CheckerConfig (BackupPath, LogPath, LockDirPath, LockFilePath, MinDiskPrimaryGB, SafetyFactor, MaxLockAge), call Validate(), and register a cleanup to release the backup lock. This makes the tests self-contained and avoids reliance on a repository lock file, reducing flakiness.

Update atomic restore behavior to propagate ownership/permissions from the nearest existing parent when creating directories or writing files. ensureDirExistsWithInheritedMeta now walks up to find an existing ancestor, creates missing nested dirs, and applies inherited uid/gid and mode reliably (with proper error handling and closes). desiredOwnershipForAtomicWrite was adjusted to repair files that have root:root by inheriting the parent's group, prefer parent gid when appropriate, and handle existing files consistently. Added unit tests to cover repairing root group on existing dest and creating nested dirs with inherited metadata. Docs updated to note that staged restores enforce ownership/permissions for created parent directories and mention the repair of common root:root group regressions.

Add a default branch in discoverRcloneBackups to cancel and skip unsupported switch cases.

Enhance writeFileAtomic to ensure durability: call f.Sync() after writing, introduce syncDir to open and fsync the parent directory (handling EINVAL/ENOTSUP as non-fatal for dir fsync), and return proper wrapped errors on failures. This ensures both file data and directory metadata are flushed for atomic writes.

Introduce test hooks (atomicGeteuid, atomicFileChown, atomicFileChmod, atomicFileSync) and refactor fs_atomic to use them instead of direct os/file calls to improve testability. Update tests to use a statOverrideFS and fakeFileInfo to simulate file metadata and capture chown calls, removing the need to run tests as root. Adjust expectations for seeded modes and add a new test to verify chown behavior when simulating root for nested directory creation.